The Effect of Breakfast in the Classroom

on Obesity and Academic Performance:

Evidence from New York City

________________________________________________________

Working Paper #02-15

May 2015

Sean P. Corcoran

New York University

Brian Elbel New York University

Amy Ellen Schwartz

New York University

2 | P a g e

This project was supported by Award Number 1R01HD070739 from the Eunice Kennedy

Shriver National Institute of Child Health & Human Development. The content is solely the

responsibility of the authors and does not necessarily represent the official views of the Eunice

Kennedy Shriver National Institute of Child Health & Human Development or the National

Institutes of Health. We would like to thank the New York City Department of Education, and

especially the Office of SchoolFood, for providing necessary data and support. Stephen O'Brien,

Armando Taddei and Dennis Barrett at the Office of School Food and Melissa Pflugh Prescott at

NYU have been particularly valuable resources to this project. Siddhartha Aneja, Michele

Leardo, and Elizabeth Debraggio provided outstanding research assistance.

Contact author: Sean P. Corcoran, Institute for Education and Social Policy, New York

University, 665 Broadway, Suite 805, New York, NY 10012. Phone: 212-992-9468. Email:

sean.corcoran@nyu.edu.

Editorial Board

The editorial board of the IESP working paper series is comprised of senior

research staff and affiliated faculty members from NYU's Steinhardt School of

Culture, Education, and Human Development and Robert F. Wagner School of

Public Service. The following individuals currently serve on our Editorial Board:

Amy Ellen Schwartz

Director, IESP

Syracuse University, NYU Steinhardt and NYU Wagner

Sean Corcoran

Associate Director, IESP

NYU Steinhardt

Leslie Santee Siskin

NYU Steinhardt

Leanna Stiefel

Associate Director, IESP

NYU Wagner and NYU Steinhardt

3 | P a g e

Abstract

Participation in the federally-subsidized school breakfast program often falls well below its

lunchtime counterpart. To increase take-up, many districts have implemented Breakfast in the

Classroom (BIC), offering breakfast directly to students at the start of the school day. Beyond

increasing participation, advocates claim BIC improves academic performance, attendance, and

engagement. Others caution BIC has deleterious effects on child weight. We use the

implementation of BIC in NYC to estimate its impact on meals program participation, BMI,

achievement, and attendance. While we find large effects on participation, our findings provide

no evidence of hoped-for gains in academic performance, nor of feared increases in obesity. The

policy case for BIC will depend upon reductions in hunger and food insecurity for disadvantaged

children, or its longer-term effects.

4 | P a g e

I. Introduction

The federal School Breakfast Program (SBP) has subsidized breakfasts for needy

children since 1966, with the aims of reducing food insecurity, improving nutrition, and

facilitating learning (Bhattacharya, Currie, & Haider, 2006; Millimet, Tchernis, & Husain, 2010;

Poppendieck, 2010). Participation in the SBP, however, typically falls well below that of its

lunchtime counterpart (Bartfeld & Kim, 2010; Basch, 2011; Dahl & Scholz, 2011; Schanzenbach

& Zaki, 2014). In New York City, for example, less than a third of all students take a breakfast

each day, even though it has been offered free to all students since 2003 and roughly three in

four students live in low income households (Leos-Urbel et al., 2013).1

To increase participation in the SBP, a number of school districts have adopted Breakfast

in the Classroom (BIC), a program that offers free breakfast to students in the classroom at the

start of the school day, rather than providing it in the cafeteria before school. The intent is to

reach students unable or unwilling to arrive early to school, and to reduce stigma associated with

visiting the cafeteria before school for a subsidized meal. NYC schools began implementing BIC

in 2007 and today the program is offered in nearly 300 of the city’s 1,700 schools.2

Advocates argue that moving breakfast from the cafeteria to the classroom provides

myriad benefits, including improved academic performance, attendance, and engagement, in

addition to reducing food insecurity among disadvantaged children. Indeed, there is robust

evidence that the consumption, timing, and nutritional quality of breakfast all affect cognitive

performance (e.g., Wesnes et al., 2003; Rampersaud et al., 2005). While there has been less work

evaluating BIC in particular, at least one study found that moving breakfast to the classroom can

1 A 2012 report from the Food Research and Action Center rated NYC last out of 26 urban school districts in

breakfast participation among subsidy-eligible students (FRAC, 2012). 2 “It’s a Hit: Breakfast in the Classroom,” The New York Times, November 17, 2008, A21.

5 | P a g e

substantially improve math and reading performance (Imberman & Kugler, 2014).3 At the same

time, others have warned BIC will have deleterious effects on students’ weight, increasing BMI

and obesity, as participants consume more daily calories or less nutritious food than they

otherwise would. In NYC, the Bloomberg administration temporarily halted the expansion of

BIC when an internal study found BIC students were more likely to eat two breakfasts, one at

home and another during school (Van Wye et al., 2013).4 There is, however, scant research

available to guide policymakers in resolving these conflicting claims, and virtually no evidence

on the impact on BMI or obesity in particular.

In this paper, we use the staggered implementation of BIC in NYC together with richly

detailed longitudinal data on student height, weight, achievement, and attendance to estimate the

program’s impact on body mass index (BMI), obesity, academic performance and attendance.

We begin by investigating whether BIC had a significant impact on schools’ average daily

participation in the breakfast and lunch programs. Next, we use longitudinal data on students to

estimate the impact of BIC on BMI and other student outcomes. This analysis uses a difference-

in-difference design, contrasting observationally similar students in schools that did and did not

adopt BIC, before and after implementation. We also estimate impacts using an event study

specification, including a series of indicators identifying years before and after BIC adoption, to

capture potential differences in trajectories prior to adoption.5 The estimated effects are

interpreted as “intent-to-treat” effects, as the treatment here is the offer of BIC to all or some

students in the school. As in most studies, we do not observe individual student meal

3 A second, unpublished paper found similar results for test scores in a subset of schools in San Diego (Dotter,

2012), although two more recent studies did not (Anzman-Frasca et al., 2015; Schanzenbach and Zaki, 2014). These

studies are described below. 4 “Hiccup in the Most Important Meal,” The New York Times, April 19, 2012, A1.

5 We provide graphical evidence that outcomes in BIC and non-BIC schools were on similar trajectories prior to

schools’ adoption of BIC, a necessary condition for the difference-in-difference specification to provide unbiased

estimates of causal effects.

6 | P a g e

consumption or classroom level participation in BIC. To address the partial adoption of BIC in

some schools, we estimate models using three increasingly stringent measures of program

coverage. Intent-to-treat effects are arguably of most interest to policymakers weighing the

benefits and costs of offering BIC, in the types of schools that choose to adopt it.

We find that NYC schools offering BIC saw a substantial increase in school breakfast

participation with no spillover effects on lunch participation. Further, there is no evidence BIC

increased BMI or the incidence of obesity among affected students. We find insignificant effects

on reading and math achievement in grades 4-8, a sharp contrast with Imberman and Kugler

(2014). Finally, we find small positive effects of BIC on attendance rates, concentrated in middle

school. While our data do not permit us to examine impacts on individual student eating

behaviors, our findings are consistent with recent experimental evidence showing BIC has, at

best, small effects on net breakfast consumption and nutrition (Schanzenbach & Zaki, 2014).

This study has important implications for the current policy debate over providing

breakfast in the classroom. While BIC promises a way to address food insecurity and hunger

among poor children, and potentially raise achievement, enthusiasm for the program has been

tempered by concerns over potential weight gain and increases in obesity. Unfortunately, there

has been little to no research on BIC’s impact on obesity, leaving policymakers and school

leaders to make decisions without the benefit of evidence. This paper aims to close that gap.

Our analysis comes from New York City, which is the largest provider of school meals in

the country and frequently regarded as a national leader in school food policy. Nearly 200

elementary and middle schools in NYC adopted BIC between 2007-08 and 2011-12, serving

30,000 students each day. Using detailed student and school-level data from this period, we

found that BIC’s significant impact on SBP participation had no deleterious effects on obesity or

7 | P a g e

weight gain. NYC’s experience suggests BIC may be a practical way for urban school districts to

reduce hunger and food security, without adverse consequences for childhood obesity.

II. Background

a. The Effects of School Meals on Health and academic

achievement

There is considerable evidence that the availability and quality of school meals programs

can affect the nutritional intake and academic outcomes of participating students. For

example, in a study of the SBP, Bhattacharya, Currie, and Haider (2006) used detailed survey

data from the NHANES III to investigate how access to the SBP affected children’s breakfast

consumption and nutrient intake. They found no impact of the SBP on total calories

consumed or the likelihood of eating breakfast, but found large effects on the nutritional

quality of breakfasts eaten, with fewer calories from fat, and higher serum levels of vitamins

C, E, and folate. Schanzenbach (2009) examined the body weight of students participating in

the school lunch program and found that children eating school lunches were more likely to

be obese than those bringing their own lunch, a finding she attributed to higher caloric intake

among students taking school lunches. A study by Millimet, Tchernis, and Husain (2010)

corroborated this finding for school lunches, but—consistent with Bhattacharya, Currie, and

Haider (2006)—found that participation in the SBP was associated with lower rates of

obesity.

Evidence of a causal impact of school meals on educational outcomes is mixed, but

frequently positive. In one study of the long-run effects of the school lunch program,

Hinrichs (2010) found sizable effects on the educational attainment of adults who were

8 | P a g e

exposed to the program early in life. In contrast, using administrative data from Chile,

McEwan (2013) found no short-run effects on test scores, school attendance, and grade

repetition of providing free high-calorie meals to poor children. Along the same lines,

Dunifon and Kowaleski-Jones (2003) found little association between school lunch program

participation in the U.S. and achievement after accounting for selection into the program.6 In

a clever study examining schools’ responses to test-based accountability in Virginia, Figlio

and Winicki (2005) found that schools under accountability pressure substantially increased

the caloric content of their meals on test days, and saw larger increases in passing rates as a

result. Consistent with this type of short-run effect, Imberman and Kugler (2014) found the

introduction of BIC into a large urban school district had large positive effects on reading

and math achievement, even when the program was implemented a short time before the test.

(We describe this study in greater detail in Section 2.2).

That the consumption and quality of breakfast can have at least a short-run effect on child

cognitive performance is confirmed in a number of experimental studies.7 For example, a

study in the U.K. randomly assigned 10-year-old students to different breakfast regimens at

home and found students receiving a higher-energy breakfast scored higher on tests of

creativity and number checking (Wyon et al., 1997). They were also less likely to report

feeling bad or hungry. Wesnes et al. (2003) randomly assigned students to receive one of

four types of breakfast on successive days (one of two types of cereal, a glucose drink, or no

breakfast) and found that students eating a cereal breakfast performed better on a series of

tests of attention and memory over the course of the morning. Simeon and Grantham-

6 More comprehensive reviews of this literature can be found in Briefel et al. (1999), Hoyland, Dye, and Lawton

(2009), Ponza et al. (1999), and Rampersaud et al. (2005). 7A more thorough review can be found in Pollitt and Matthews (1998) and Hoyland, Dye, and Lawton (2009).

9 | P a g e

McGregor (1989) conducted a small experiment in which under-nourished children in the

West Indies were randomly assigned to receive a breakfast or a cup of tea on alternate days.

After consuming breakfast, students performed better on cognitive tests of arithmetic and

problem solving than when drinking only tea.

Relevant to BIC, one study we identified found that when breakfast is consumed relates

to its effects on cognitive performance. In a randomized control trial, Vaisman et al. (1996)

found that 11- to 13-year-old students who ate a regular breakfast before school (two hours

before testing) performed no better than a control group on tests of cognitive functioning.

However, students who ate a cereal and milk breakfast in class 30 minutes before testing

performed significantly better.

b. Breakfast in the classroom

Breakfast in the Classroom alters the traditional SBP by offering a free breakfast in class

at the start of the school day, rather than in the cafeteria before school hours (FRAC, 2012).

The intent is to increase breakfast participation among students who are unable or unwilling

to arrive early to school, and to reduce stigma associated with visiting the cafeteria before

school for a subsidized meal. BIC advocates have argued the program also provides an

opportunity to integrate nutrition into the curriculum, as teachers can use the time to teach

good eating habits. Proponents tout the social aspects of the program as well, citing the

benefits of communal eating.8

BIC breakfasts are typically offered during the first 10-20 minutes of class. Meals are

often bagged the prior evening by school food staff, placed into insulated containers, and

8See http://www.healthyeating.org/Schools/Tips-Trends/Article-Viewer/Article/142/breakfast-at-school-a-

communal-meal-that-makes-a-difference.aspx.

10 | P a g e

refrigerated overnight. They are then delivered to classrooms in the morning, or distributed to

students as they arrive (“Grab and Go”). Because breakfasts are typically assembled the night

before, BIC menus generally differ from those prepared in the cafeteria. Specifically, BIC

meals usually consist of cold, pre-packed items such as cereal, fresh fruit, or bagels. Cafeteria

breakfasts, on the other hand, include hot meals such as pancakes or egg omelets.9 Though

they may differ in menu offerings, BIC meals are required to meet the same federal

nutritional guidelines as cafeteria breakfasts. For federal reimbursement purposes, they are

accounted for in the same way as traditional cafeteria breakfasts.

The Food Research and Action Center (FRAC), a national advocate for the SBP in

general and BIC in particular, credits BIC for high rates of SBP participation in urban school

districts like Detroit, Houston, Newark, San Antonio, Washington, D.C., and Providence

(FRAC, 2012). There has been little research, however, on BIC’s effects on breakfast

participation, academic or behavioral outcomes, or weight. One evaluation of a 2003-04 BIC

pilot in upstate New York found SBP participation doubled after implementation of the

program, and found modest improvements in attendance, behavior, and tardiness (Murphy,

Drake, & Weineke, 2005). The study, however, lacked a control group and involved only a

small number of schools. Anzman-Frasca et al. (2015) examined the effects of BIC on SBP

participation in a large urban school district and estimated that BIC increased participation by

30 percentage points, relative to matched non-BIC schools. Similarly, in re-analyzing data

from an earlier U.S. Department of Agriculture experiment, Schanzenbach and Zaki (2014)

found that BIC had a large positive effect on SBP participation and increased the likelihood

9 Not all schools in NYC have kitchen facilities for preparing hot meals. These schools would likely have served

cold breakfasts prior to adopting BIC. Schools with kitchens may have offered cold items as menu options.

11 | P a g e

that students ate a “nutritionally substantive” breakfast. They found the effect of BIC on SBP

participation was considerably larger than the offer of free breakfast alone.

Imberman and Kugler (2014) provide strong quasi-experimental evidence on the effects

of offering BIC on achievement. That study examined the impact of BIC on math and

reading scores in 5th

grade, and attendance and report card marks in grades 1-5. Their setting

was an urban district that—like NYC—had previously offered free breakfast to all students,

regardless of subsidy eligibility (Leos-Urbel et al., 2013). The district implemented BIC in 85

schools over a period of 11 weeks in 2009-10, which enabled the authors to contrast

outcomes in early adopter schools (those implementing BIC before the test) with those in late

adopting schools (implementing after the test). They found substantial intent-to-treat effects

of BIC on reading and math achievement (0.10 s.d.), with larger effects for initially low-

achieving students (0.13 – 0.14 s.d.), Hispanics (0.14 – 0.15 s.d.), and low-BMI students

(0.26 s.d.). As many students already participated in the SBP, the treatment effect of BIC on

the treated in this district was potentially much higher. They found no impact of BIC on

attendance rates or report card grades.

Interestingly, the achievement effects found in Imberman and Kugler did not vary with

the amount of time students had been offered BIC. Thus, even schools that adopted BIC as

little as one week prior to testing experienced gains in test performance. While seemingly

implausible at first, their finding mirrors Figlio and Winicki (2005), who found that

increasing caloric content of lunch on test day can improve test performance. It also aligns

with the experimental evidence described in Section 2.1 on the short-run effects of breakfast

consumption and content on cognitive performance. Imberman and Kugler’s finding of no

impact on grades but a large impact on test scores supports a short-run caloric effect and lack

12 | P a g e

of a sustained, long-run impact on achievement, but the study’s short duration makes it

difficult to rule out long-run effects.

Like Imberman and Kugler (2014), an unpublished study by Dotter (2012) used the

introduction of BIC in San Diego over a 4-year period to estimate its effects on achievement

in 2nd

– 6th

grade, attendance, and classroom behavior. In contrast to NYC and the district in

Imberman and Kugler (2014), San Diego had only previously offered universal free

breakfasts in schools with Provision 2 status under the National School Lunch Act (“UFM

schools”).10

All others offered breakfast free or at a reduced rate to subsidy-eligible students

and at full price to other students. BIC in San Diego thus coincided with a shift to universal

free breakfast in schools that were not already UFM, making it difficult to disentangle the

BIC and price effects. Like Imberman and Kugler, Dotter found large effects of BIC on

achievement (0.11 s.d. in reading and 0.15 s.d. in math), but only in schools that did not

already offer free breakfast to all students. Thus, the effect should be interpreted as the

combined effect of BIC and free breakfast. He found no effect on attendance, but large

positive effects on teacher-reported classroom behavior, such as exhibiting “respect for

people and property.”

Finally, a third quasi-experimental study by Anzman-Frasca et al. (2015) examined the

impact of BIC on meals program participation and achievement in a large urban school

district in 2012-13. They found a large effect of the program on SBP participation (noted

above), but unlike the other two studies described here, found no impact on student

10

Under Provision 2, a school may certify children as eligible for free or reduced-price meals for up to four

consecutive years—without collecting annual data on eligibility—and provide meals free of charge to all students.

The intent is to reduce the administrative burden on schools and parents related to proving income eligibility. For

details see: http://www.gpo.gov/fdsys/pkg/CFR-2012-title7-vol4/pdf/CFR-2012-title7-vol4-part245.pdf [last

accessed May 15, 2015].

13 | P a g e

achievement in 2nd

-6th

grade. Anzman-Frasca et al. did find a positive, statistically significant

effect of BIC on attendance, but the effect size was very small.

These three studies offer strong evidence on the academic impacts of BIC, and go far

beyond what was previously available. However, they have several limitations. First, none

provided evidence of the program’s effect on student weight, an important outstanding

question in the literature. Second, two of the three relied on relatively small samples of

elementary schools. Imberman and Kugler’s main estimates use 5th

graders in approximately

85 treatment and 19 control schools; Dotter looked at a broader range of grade levels, but a

smaller number of schools (45 treatment and 22 control). His sample of non-UFM schools—

where the only significant effects were found—was smaller (19 treatment and 16 control).

Third, only Dotter is able to say much about long-run effects, up to four years after the first

BIC implementation. Imberman and Kugler provide a clean estimate of the short-run impact

of BIC, but their results may say more about calorie intake and the short-term malleability of

test performance than about the long-run consequences of adjusting children’s eating habits.

Our analysis improves on prior work by incorporating annual student-level measures of BMI

and a larger sample of students and BIC schools at both the elementary and middle school

level.11

Three studies that we are aware of have looked at the relationship between BIC and

calorie consumption or weight, although only one has a design that can support causal

claims. Baxter et al. (2010) collected cross-sectional data on BMI, breakfast program

participation, and energy intake (from researcher observations) for a sample of 4th

grade

11

Some high schools in NYC have implemented BIC. However, as we explain in the next section, our BMI data on

high school students and schools appears to be less reliable than our data for elementary and middle school. Hence

our analysis in this paper is restricted to students in elementary and middle grades.

14 | P a g e

students in 17 schools, seven of which had adopted BIC. They found children in BIC

consumed more calories at breakfast and had higher BMIs, on average, than children eating

breakfast in the cafeteria. Van Wye et al. (2013) surveyed students in nine BIC and seven

comparison schools in NYC and found students offered BIC were more likely to eat more

than once in the morning, consuming 95 more calories, on average, than students in schools

not offering BIC. Access to student-specific data on food consumption is an obvious

advantage of these studies. However, both were correlational with a small number of schools

that did not address selection into BIC. Schanzenbach and Zaki (2014) provide stronger

evidence of the program’s effects on breakfast consumption and obesity by re-analyzing data

from an experiment conducted by the U.S. Department of Agriculture in the early 2000s

(Bernstein et al., 2004). In the original experiment, treatment schools offered universal free

breakfast and could choose to serve it in the cafeteria or in the classroom. In a comparison of

BIC and control schools that did not offer free breakfast, the authors found large effects of

BIC on participation, and a five percentage point increase in the likelihood of eating more

than one breakfast, but modest effects on breakfast consumption, and no effect on BMI or

nutrition. They conclude that—for most students—BIC changed the location and timing of

breakfast rather than the propensity to consume it. Not surprisingly in light of this, they find

BIC had no impact on achievement or behavior. While randomization of free breakfast is an

important feature of this study, it is worth noting the decision to offer BIC was not random,

making the comparison of BIC and other treated schools (offering free breakfast only) non-

experimental. Further, the number of BIC schools (18) was small relative to the cafeteria

group (61). The community context for schools in the USDA study was also quite different

from ours (NYC).12

Thus, there remain open questions about the efficacy of BIC in the

12

The participating school districts included Boise, ID; Shelby County, AL; Harrison County, MS; Phoenix, AZ;

15 | P a g e

context of universal free breakfast in the cafeteria—the context increasingly relevant for

urban school districts nationwide.

III. Data

a. data Sources, Outcomes, and sample definitions

We draw on four primary data sources, all provided by the New York City Department of

Education (NYCDOE) and its Office of School Food. The first is a database of BIC

participation that includes start dates for schools that ever adopted BIC, and the extent of

program coverage in the school (e.g., grades served and the number of BIC and total

classrooms). The second is longitudinal school-level data on breakfast and lunch

participation for all regular public schools that served elementary and/or middle grades

between 2001-02 and 2011-12. This data provides annual counts of meals served, average

daily attendance (ADA), and Provision 2 (UFM) status, and includes 1,000 to 1,100 schools

enrolling 713,000 to 730,000 students, depending on the year. The third is administrative

data for the universe of students in NYC public schools between 2006-07 and 2011-12,

including demographics, educational needs and program participation, standardized test

scores, and attendance rates. Finally, the fourth is annual student height and weight

measurements collected through the city's Fitnessgram program; these are used to compute

BMI and indicators of obesity, as described below. We exclude high schools, private, charter,

prekindergarten, alternative, and other special schools and programs from the study.13

Santa Rosa, CA; and Wichita, KS. 13

Conversations with the NYCDOE and the NYC Department of Health and Mental Health raised concerns about

the quality of Fitnessgram data in high school. Whereas height and weight measurements in elementary and middle

schools are taken by school staff (e.g., the school nurse or P.E. teacher), they are frequently self-reported in high

16 | P a g e

The student-level administrative data tracks students longitudinally as they progress

through school. These data include school of record, state test scores in English Language

Arts (ELA) and mathematics (both in grades 3-8 only), gender, race/ethnicity, age, household

income eligibility for free or reduced price meals, recent immigrant status, days in

attendance, days enrolled, and participation in other educational programs (e.g., special

education and/or ELL). We standardized ELA and math scores within grade, subject, and

testing year. Grade 3 test scores are used only as a lagged outcome for students in 4th

grade.

We calculated student attendance rates as the number of days present as a percentage of days

enrolled.

NYC schools have conducted the Fitnessgram since 2005-06 as part of the district's

standards-based physical education program.14

Fitnessgram requires all schools to collect

students’ height and weight annually, and to assess students’ aerobic fitness, muscle strength,

endurance, and flexibility. At the end of the year, students receive a report that summarizes

their performance and suggests ways for them to reach their “Healthy Fitness Zone,” targets

for better health based on their age and gender. While school coverage rates were lower in

the early years of the Fitnessgram, by 2012 nearly 1,700 schools were participating each

year, providing data on more than 860,000 students in all grades.

From the Fitnessgram we used students’ weight (in pounds) and height (in inches) to

compute the standard BMI measure, (weight/height2)*703. Biologically implausible values—

defined as more than four s.d. below or five s.d. above the mean for the students’ gender and

age in months—were set to missing. Following the CDC definition of child obesity, we

school. Our concern about measurement error in the high school data led us to exclude these students from our

analysis. 14

See Rundle et al. (2012) and Elbel et al. (2013).

17 | P a g e

classified children as obese if their BMI was at or above the 95th

percentile nationally for

their gender and age in months.15

Though there is some debate in the public health literature

over the best measure of adiposity in children (e.g., Cole et al., 2005; Mooney, Baecker, &

Rundle, 2013), BMI is the most common. Our analyses of student weight use two outcome

measures: BMI standardized by gender and age, using all students in the Fitnessgram (z-

BMI), and a 0-1 indicator for BMI above the CDC obesity threshold.

Our analytic samples for BMI/obesity, achievement and attendance overlap, but differ in

two key ways. First, the included grade levels vary, depending on data availability. Models

for BMI and attendance include students in grades K-8, while the models for achievement

include only grades 4-8. Second, the BMI sample excludes students in school-years where

Fitnessgram participation was lower than 50 percent. These students were not excluded from

the other models. In allowing these analytic samples to differ, we make full use of available

data. (The online appendix reports estimates from models estimated using a fixed sample of

students. The results are very similar).

b. BIC Adoption in New York city

Figure 1 shows the cumulative number of NYC schools that adopted BIC, by month,

between 2007-08 and 2011-12. (The figure is limited to schools enrolling grades K-8, and

excludes private and charter schools). The largest number of adoptions occurred in early

2010-11, although a significant number of schools began offering BIC in 2008-09 and 2009-

10. By the end of 2011-12, 195 elementary and middle schools had adopted the program,

15

CDC percentiles are based on the national distribution of BMI in 2000-01, the most recent growth charts in use.

See http://www.cdc.gov/growthcharts/clinical_charts.htm.

18 | P a g e

with an average daily participation of more than 30,000 students.16

It is worth noting that

some schools adopted BIC in the middle of the school year, suggesting its full effect on

annually measured outcomes may not be observed until the following year.

Not every school adopted BIC school-wide, which has implications for how we define

“treatment.” In some cases, BIC was offered only to specific grades or a subset of classrooms

within the school.17

In the absence of data linking specific students or classrooms to BIC, we

used information on the number of BIC and total classrooms, BIC grades served, and school

grade span to create three increasingly stringent indicators of program coverage in a school.18

The first indicator is equal to one for schools offering BIC in any classroom (and zero

otherwise). The second is equal to one for schools offering BIC in at least 25% of its

classrooms, but not school-wide, which eliminates schools piloting the program with a small

number of students. Finally, the third is equal to one for schools offering BIC school-wide

(“full BIC”).19

The bars in Figure 1 show the cumulative number of schools that adopted BIC

according to these definitions. While the majority of schools fall only in the first category, by

the end of 2011-12, 62 elementary and middle schools offered BIC to at least 25% of

classrooms, and 23 offered it school-wide.

16

Including regular high schools, the total number of BIC schools in 2011-12 was 279 schools with an average daily

BIC participation of 36,000. Adding private, charter, pre-kindergarten, alternative, and special education schools the

total number of schools was 348 with an average daily participation of 41,500. 17

In NYC, as in other large districts, the specific timing of program changes—whether in school food or

otherwise—reflects the interplay of opportunity, convenience, capacity, politics, and economics. In the case of BIC,

our conversations with the Office of School Food and school personnel suggest that timing depended, in part, on the

physical layout of the school, which determines the logistics for delivering breakfast to classrooms. A similar

process likely plays out within schools in determining which classrooms offer BIC. 18

BIC adoption is based on data collected by the NYC Office of School Food in November 2009, March 2011, and

October 2012. Each wave provided contemporaneous information about BIC participation and start date. We used

information from all three waves to construct our treatment variables. 19

We are less confident in the reported data on specific grades offering BIC in each school, where applicable, and in

the exact percentage of classrooms offering BIC. This led us to the broader classification described here. We have,

however, estimated models using these alternative measures of treatment (available in the online appendix). The

results are qualitatively the same.

19 | P a g e

Table 1 reports baseline mean characteristics for BIC schools and schools that never

adopted BIC. (The BIC categories are not mutually exclusive, and represent our three

increasing levels of program coverage). On a number of dimensions, BIC schools were more

disadvantaged than those that never adopted the program. For example, they enrolled a

greater percentage of students eligible for free meals (74.1% versus 67.3%) and had higher

concentrations of black, Hispanic and ELL students. Standardized test scores were lower on

average in BIC schools, both in reading (-0.13 s.d.) and in math (-0.16 s.d.), and were lower

still in schools with more extensive BIC coverage. Schools in all five boroughs adopted BIC,

though they were over-represented in the Bronx. Average daily participation in the school

breakfast program was slightly lower at baseline in schools that adopted BIC versus those

that did not (23.3% vs. 24.5%). Finally, students in BIC schools on average had higher BMI

at baseline than those attending non-BIC schools, especially in schools with greater BIC

coverage. Roughly one in four students were obese in schools that adopted BIC school-wide,

compared with roughly one in five in non-BIC schools.

IV. Empirical strategy

Our main empirical models are difference-in-differences regressions at the school or

student level (depending on the outcome) which contrast outcomes in schools that did and

did not adopt BIC, before and after implementation. Included covariates and school fixed

effects account for non-random selection of schools into the program on the basis of

observed characteristics or time-invariant factors correlated with the outcome of interest

(e.g., meals program participation, BMI, achievement, attendance). We address the

possibility that BIC and non-BIC schools were on different trajectories prior to schools’

20 | P a g e

adoption of BIC through event study models, described below and shown graphically in

Figures 2-3.

We estimate the impact of BIC on meals program participation using annual data at the

school level. The following model is estimated for breakfast and lunch participation, for all

schools combined and separately for elementary and middle schools:

(1) (𝐴𝐷𝑃𝑚𝑠𝑡

𝐴𝐷𝐴𝑠𝑡∗ 100) = 𝛿 ∗ 𝐵𝐼𝐶𝑠𝑡 + 𝛽′𝑾𝑠𝑡 + 𝛼𝑡 + 𝛾𝑠 + 𝑣𝑠𝑡

In (1), the dependent variable is the average daily participation rate for meal m (breakfast

or lunch) in school s and year t, defined as the average number of m meals served per school

day (ADPm) divided by average daily attendance (ADA) and multiplied by 100. This can be

interpreted as the percentage of students in attendance who take a meal m on an average day

in school s in year t. BICst is equal to one if school s adopted BIC prior to the end of school

year t (and zero otherwise), and implemented alternately using our three categories of

program coverage. γs and αt are school and year effects, respectively, and Wst is a vector of

school-level covariates, including total enrollment, percent female, percent by race/ethnicity,

percent ELL, percent in special education, and percent eligible for free or reduced price

meals. Standard errors are adjusted for clustering by school. In alternative specifications of

(1), we add school-specific linear time trends to allow for differential trends over time in the

outcomes.

We estimate the impact of BIC on student outcomes using annual data at the student

level, again estimating separate models for elementary and middle school students. (The

reason for this, as we show in section 5.1, is that the impact of BIC on breakfast participation

is proportionally much larger in middle schools, where baseline participation was much

21 | P a g e

lower—perhaps due to stigma.) Each model takes the following form for outcome Yigst (BMI,

obesity, math score, reading score, and attendance) for student i in grade g, school s, and year

t:

(2) 𝑌𝑖𝑔𝑠𝑡 = 𝛿 ∗ 𝐵𝐼𝐶𝑖𝑠𝑡 + 𝛽′𝑿𝑖𝑡 + 𝛼𝑡 + 𝛾𝑠 + 𝑢𝑖𝑡

BICist, γs and αt are defined in the same manner as in (1), and Xit is a vector of student

covariates potentially related to both the outcome and the adoption of BIC in school s. These

include eligibility for free or reduced price meals, race and ethnicity, limited English proficiency,

special education status, age in months (as of the date of BMI measurement and in the

Fitnessgram models only), and lagged math or reading scores (in the math and reading models,

respectively).20

uit is a student-year error term, and standard errors are adjusted for clustering by

school. As before, BIC is defined alternately using the three categories of coverage defined

earlier. In a second set of specifications, we allow the effect of BIC to vary by the extent of time

the student has attended a BIC school, with BICist equal to the cumulative number of days, in

hundreds, that student i has attended any school offering BIC prior to the date their outcome was

measured in year t.21

To test the identifying assumption that outcomes in BIC and non-BIC schools were on

similar trajectories prior to adoption of the program, we estimate “event study” versions of

models (1) and (2). In these regression models, we replace the BIC treatment variable with a set

20

A single variable in the model indicates whether the student is eligible for free or reduced price meals, or whether

the student is enrolled in a UFM school. In cases where eligibility status is missing, this variable is equal to zero. We

include an additional indicator variable equal to one for students with missing eligibility information. 21

This treatment is intended to capture heterogeneity in student exposure to BIC, and—like the school-level

treatment—is implemented using the three increasingly stringent thresholds of BIC coverage. To the extent BIC

takes time to have an effect on BMI and/or achievement, this specification may be more appropriate. We have

experimented with numerous variations on these treatment definitions. These alternatives are described later in the

paper, where appropriate. Estimates from these alternative approaches are available in the online appendix.

22 | P a g e

of indicators corresponding to specific years before and after BIC adoption. (These indicator

variables are always zero for schools that never adopted BIC, and the reference period for BIC

schools is the first year of the program). All other covariates from (1) and (2) are included in

these models. Any significant differences in mean outcomes that we observe in the pre-treatment

period could reflect differential trends in BIC schools prior to adoption of the program.

Differences in mean outcomes in the post-treatment period can be used to test for differential

effects over time (as in Dotter, 2012).

V. Results

a. The impact of bic on school meals program participation

Table 2 reports estimates of the impact of BIC on average daily school breakfast and

lunch program participation from model (1). Each cell in this table is a coefficient and

standard error estimate from a separate regression, for the treatment definition shown in the

row and the school sample shown in the column. Rows (1)-(3) are impact estimates for

breakfast participation and rows (4)-(5) are estimates for lunch participation. All models

include school and year fixed effects, as well as school-level covariates.

The point estimates in Table 2 show a large increase in breakfast program participation in

schools offering BIC, with no corresponding change in lunch participation. Across all

schools serving elementary and middle grades, we find BIC increased SBP participation rates

by 11.8 percentage points, on average, from a baseline of 20.1 percent—a nearly 60 percent

increase. Although point estimates from separate elementary and middle school regressions

are an identical 13 percentage points, the effect size in middle school is proportionately much

larger, since these schools had lower baseline participation rates (12 percent versus 23.6 in

23 | P a g e

elementary schools).22

As expected, we observe larger effects on participation rates when

BIC classroom coverage is higher. For example, the estimated impact on average daily

breakfast participation is 21 to 25 percentage points when at least 25 percent of classrooms

offered BIC, and 29 to 32 percentage points when BIC was offered school-wide. (All

coefficient estimates for breakfast program participation are statistically significant at the

0.01 level or better). This finding provides support for our three-tiered classifications of BIC

coverage at the school level, as well as a strong “first stage” effect of BIC on program take-

up that would be needed if BIC were to have an effect on other student-level outcomes.23

In contrast, we find that offering BIC had no significant impact on lunch program

participation. In all cases, the coefficient estimates for lunch participation in Table 2 are

small—usually less than 1 percentage point—and are statistically insignificant. Taken

together, there is no evidence to suggest BIC crowded out lunch program participation or

encouraged greater participation (say, by reducing stigma associated with subsidized meals).

The difference-in-difference estimates are valid as causal effects only if BIC and non-

BIC schools were experiencing similar trends in meals program participation prior to the

adoption of BIC. In Figure 2 we show coefficients from event study regression models, in

which separate indicator variables are included for BIC schools in years prior to and

following BIC adoption. (Reference year zero is the year prior to adoption). For clarity, the

22

In Table 2, the count of schools appearing in the “all schools” regressions is larger than the sum of schools

appearing in the “elementary” and “middle school” only regressions, since the former includes schools serving both

elementary and middle grades. Unfortunately, meals served data is not disaggregated by grades within schools. 23

As another check on the quality of our data, we were able to obtain high-frequency point-of-sale (POS) data for

three BIC schools, which allowed us to observe how daily meal service changed with implementation of BIC. The

advantage of POS data is that it is generated by student transactions, not school counts. In two of the three schools,

we observe a substantial increase in SBP participation immediately after BIC adoption, at 36 and 53 points,

respectively. (The third school, which was demographically quite different from the other two, saw no increase).

This analysis is provided in the online appendix. While only three schools, this detailed look provides additional

confidence in the validity of the meals served data and in the BIC implementation dates.

24 | P a g e

figure shows the coefficients and confidence interval bounds for two of the three BIC

categories (>25% BIC and full school). We observe virtually no trend in regression-adjusted

participation rates prior to the adoption of BIC. The hypothesis that breakfast and lunch

participation rates were equal to those in the reference year cannot be rejected, up to seven

years prior to BIC adoption. After adoption, participation rates in BIC schools were 10-18

percentage points higher in the first year, depending on the extent of BIC coverage, and 18-

37 percentage points higher in the second year. In schools where BIC was adopted school-

wide, breakfast program participation continued to increase three and four years later. By

comparison, the event study for lunch shows no discernible change in participation rates for

any year following BIC adoption. (While the estimation sample for Figure 2 was an

unbalanced panel, repeating the analysis with a smaller balanced panel of schools produced a

nearly identical result.

Including school-specific linear time trends to the models in Table 2 had only modest

effects on these estimated impacts. For example, among all schools, the effects on breakfast

participation rates are 8.3, 17.2, and 26.4 percentage points for the three treatment intensities,

respectively. All remain statistically significant at the 0.01 level. (Results are available in the

online appendix).

b. The impact of bic on BMI AND OBESITY

Estimates of the impact of BIC on student BMI and obesity are reported in Table 3. In

this table, each cell reports the estimated BIC treatment coefficient and standard error from a

separate regression. The regressions differ in their estimation sample (grades K-5 or 6-8),

specification of the BIC treatment (pre-post indicator for school s, or cumulative days of

exposure for student i divided by 100), and extent of program coverage in the school (any

25 | P a g e

BIC; >25% BIC; full BIC). Impact estimates for BMI are given in panel A, while estimates

for obesity are in panel B. Results in later subsections are organized in an analogous manner.

We find no evidence that the offer of BIC increased BMI or the prevalence of obesity.

Many of the point estimates in Table 3 are negative, which if significant would indicate that

students have lower BMI and lower rates of obesity when their school offers BIC than

observationally similar students attending the same schools when not offering BIC. All of the

effect sizes are small, however, and none of the effects are statistically significant. Focusing

on the difference-in-difference results in the first two columns, the point estimates are largest

(in absolute value) for full BIC schools, as would be predicted if a treatment effect were

present and were more likely in a school with greater BIC coverage. However, we cannot

reject the hypothesis that the coefficients are the same across the different treatment

measures. Interestingly, when treatment is measured using cumulative days of student

exposure, the point estimates are sometimes positive—though never significant—hinting that

prolonged BIC exposure could be associated with higher BMI.

As noted, we experimented with a number of alternative specifications for BIC

treatment.24

For example, we set the cumulative days of exposure to zero if the student was

not currently in a BIC school; in most cases these point estimates were smaller in absolute

value than those in Table 3. Alternatively, we defined cumulative days of exposure as the

number of treated days prior to measurement in year t only, with a separate dummy variable

indicating BIC treatment in prior years. The point estimates were again closer to zero than in

Table 3. Finally, we estimated models in which three mutually exclusive categories of BIC

24

Results using these alternative specifications are available in the online appendix.

26 | P a g e

coverage were jointly included in the model. None of the results were materially different

from those in Table 3.

c. The impact of bic ON STUDENT ACHIEVEMENT

Table 4 reports our estimates of the impact of BIC on student achievement in ELA and

mathematics. As in Table 3, each cell is the estimated treatment effect from a separate

regression, with comparable estimation samples, treatment variable specifications, and BIC

coverage thresholds. The main differences are that the elementary school models use only

grades 4-5, since testing begins in 3rd

grade and lagged achievement is included as a

covariate.

We find no consistent evidence of an impact of BIC on achievement in either subject.

Most point estimates are small in absolute value and vary in sign, with ELA estimates mostly

negative and math estimates mixed. In ELA, we observe a statistically significant and

negative effect of -0.020 s.d. (p<0.05) for middle school students attending schools with at

least 25% classroom coverage, an effect that is larger with prolonged exposure to BIC (-

0.005 s.d. per 100 days; p<0.05). In math, the only statistically significant effect we observe

is for middle schoolers attending schools with at least 25% classrooms coverage, a positive

0.008 s.d. per 100 days (p<0.05). The point estimates are larger for students attending

schools that adopted BIC school-wide, but are not significant. With a large sample and

multiple models, it is not surprising to find some statistical significance. On balance, we see

no strong evidence of an impact of BIC on academic performance.

Here again, point estimates for the effect of BIC on achievement tend to be larger in

schools with greater BIC coverage, although this pattern is not consistent, and we cannot

27 | P a g e

reject the hypothesis that the effects are the same across categories. When BIC treatment is

measured using cumulative days of student exposure, the results suggest that prolonged BIC

participation could be associated with lower ELA scores and higher math scores. The

cumulative effects would be small, however.25

Indeed, all of the achievement effects we

observe are considerably smaller than those estimated in Imberman and Kugler (2014) and

Dotter (2012). In the concluding section of the paper, we consider several potential

explanations for the divergent findings.

We experimented with the same alternative specifications of the BIC treatment variable

for our achievement models as we did for BMI and obesity. In no case were the results

materially different.26

Additionally, because the estimation sample differs for the

BMI/obesity and achievement outcomes, we estimated both sets of models with a fixed,

identical sample of students with sufficient data to be included in all models.27

Again, the

results were qualitatively very similar.

d. The impact of bic ON attendance

Table 5 reports estimates of the impact of BIC on attendance rates, measured as the

number of days present as a percent of days enrolled. As in earlier tables, each cell is the

estimated treatment effect from a separate regression, with comparable estimation samples,

treatment variable specifications, and coverage thresholds. For the attendance outcome, the

elementary school models include all grades K-5 (not just the tested sample). The middle

school models continue to use grades 6-8.

25

For example, a middle school student 1 s.d. above the mean in cumulative days in a school with at least 25%

classroom coverage would be predicted to have 0.006 s.d. lower ELA scores and 0.009 s.d. higher math scores. 26

Results using these alternative specifications are available in the online appendix. 27

For instance, students in grades K-3 contribute to the BMI/obesity and attendance estimates in Tables 3 and 5 but

not to the achievement estimates in Table 4, since they lack a current and lagged test score. Moreover, some students

with test scores did not have Fitnessgram data and thus did not contribute to the BMI estimates.

28 | P a g e

In this case the estimated effects in middle school are positive, and some are statistically

significant (p<0.05). Students in middle schools offering BIC had 0.3 to 0.4 percentage point

higher attendance rates when BIC was offered in their school, depending on the extent of

BIC classroom coverage. In all cases, the estimated effect sizes are very small. In elementary

school, attendance rates are already high—about 92 percent in schools that ever adopted

BIC—and in middle school about four points lower, on average (88 percent). Assuming a

180-day school year, a 0.50 percentage point increase in attendance translates into 0.9 school

days. Thus, the average effect found here amounts to about one quarter to less than one full

school day.

In complementary work (not shown here), we looked at the potential “socializing effects”

of BIC by testing for effects on responses of middle school students to NYC’s School

Environment Survey. This survey aims to capture student attitudes toward their school,

classroom, and teachers, with items such as “I feel welcome in my school,” “Most of the

adults I see at school every day know my name or who I am,” “I am safe in my classes,” and

so on. We did not find any consistent effects of BIC on responses to any of these survey

items.

e. EVENT STUDY MODELS

As we did for school breakfast and lunch participation in section 5.1, we estimated event

study regression models for each of the student-level outcomes, school levels, and categories

of BIC coverage. The results are summarized in Figure 3. For clarity of presentation, this

figure shows only models using the intermediate category of BIC program adoption (>25%

BIC). (Figures using the other definitions of treatment are similar, and provided in the online

appendix. Year zero is again the year prior to adopting BIC.

29 | P a g e

As in Figure 2 we find no evidence of a differential trend in math and ELA achievement

in BIC schools prior to program adoption. Differences from year zero are small in magnitude

and statistically insignificant. The same holds for BMI, obesity, and attendance, especially in

grades K-5. There is somewhat of an upward trend in BMI, obesity, and attendance in grades

6-8 prior to program adoption, although differences from year zero are again statistically

insignificant. Consistent with our main results in Tables 3-5, we find no evidence that BIC

significantly impacted BMI, achievement, or attendance, with the possible exception of

increased attendance in middle school.

VI. Discussion

BIC has been widely adopted by school districts across the United States, with the goal of

increasing SBP participation and ensuring no child starts the school day hungry. Advocates

argue moving breakfast from the cafeteria to the classroom will provide myriad other

benefits as well, including improved attendance, engagement, and academic performance.

One recent study supports the latter claim, finding a substantial impact on math and reading

performance (Imberman & Kugler, 2014). Whether these effects represent short-run boosts in

test scores or sustained, long-run impacts on academics remains unclear.

At the same time, others have warned that BIC will have deleterious effects on students’

weight, increasing BMI and obesity, as participants consume more daily calories or less

nutritious food than they otherwise would. BIC expansion in NYC was temporarily

suspended over this very claim. The evidence base on BIC’s impact on obesity, however, is

thin. Our analysis using longitudinal Fitnessgram measures of student BMI indicates these

30 | P a g e

fears are largely unwarranted. We find no evidence BIC increased BMI or the incidence of

obesity among students attending schools in New York City offering the program. Nearly all

of our point estimates suggest lower average body mass in schools when students are offered

BIC, though these effects are small and not statistically different from zero.

At the same time, our study finds large positive effects of BIC on SBP participation, an

effect that did not come at the expense of lower lunch participation. We find no evidence of

an impact on reading and math achievement in schools that adopted BIC. That class time

devoted to BIC did not adversely affect achievement is encouraging. But in no case do our

estimates suggest the positive effects found in two out of the three prior quasi-experimental

studies. They do, however, align with a recent re-analysis of experimental data showing BIC

has small net effects on breakfast consumption and in turn no effect on achievement.

There are a number of reasons why our achievement effects may depart from those in

earlier quasi-experimental studies. First, the “first stage” impact of BIC on program

participation may have been weaker in NYC than in other settings. Table 2 revealed a

substantial increase in SBP take-up in BIC schools, but this increase was smaller than that

observed in San Diego (Dotter, 2012). In that study, SBP participation in BIC schools

exceeded 90 percent, well above what we observe in NYC.28

As expected, we did not

observe as large an impact on school-wide participation in NYC schools that adopted BIC in

a subset of classrooms, which may contribute to our overall null finding on achievement.

However, we also found no effects in schools that offered BIC school-wide, where the

impact on participation was much larger. Understanding variability in the effects of BIC on

28

Imberman and Kugler (2014) do not estimate the impact of BIC on breakfast participation in their district but

highlight a pilot study that documented 80 percent participation in BIC schools (versus 41 percent in all other

schools). Both rates are higher than those observed in NYC.

31 | P a g e

take-up across contexts will be an important question for future research. Second, NYC was

already offering free breakfast to students citywide prior to BIC (Leos-Urbel et al., 2013).

Dotter (2012) found effects on achievement in San Diego schools, but only in those that did

not already offer universal free breakfast. This is consistent with our findings, and may

suggest few added benefits of BIC—at least for achievement—beyond those provided by free

breakfast.

While our study improves on existing work in a number of ways, it has several

limitations. Imperfect measures at the classroom level led us to adopt three relatively broad

categories of BIC coverage aiming to capture the extent to which BIC is offered in a school.

Our categories are validated by an increasingly strong relationship with SBP participation

post-implementation, but could be improved with classroom- or individual-specific meals

data. In ongoing work, we are using newly-acquired point-of-sale (POS) data provided by the

NYC Office of School Food to track individual daily student breakfast participation for a

subset of schools.

Despite these limitations, our study is one of the first to examine the effects of offering

BIC on BMI and obesity. It uses annual student-level data on obesity from New York City,

the largest school district in the country, where nearly 300 schools adopted BIC by 2011-12

(and roughly 200 schools serving grades K-8 were used in our models). This data includes

observations on students as many as four years post-BIC implementation, allowing us to

potentially detect long-run effects. Breakfast in the Classroom has received considerable

scrutiny and media attention in NYC, and school districts nationwide have followed its roll-

out closely. Our investigation of the impact of BIC yields evidence of significant increases in

SBP participation and (small) improvements in middle school attendance, with no effect on

32 | P a g e

lunch participation, academic performance, or any weight outcome, including BMI and

obesity. Thus the modest positive impacts of BIC do not come at the cost of worsening

childhood obesity. School districts concerned about hunger and food insecurity among their

most vulnerable students might do well to consider Breakfast in the Classroom to address

these issues.

33 | P a g e

References

Anzman-Frasca, S., H.C. Djang, M.M. Halmo, P.R. Dolan, & C.D. Economos. 2015. Estimating

Impacts of a Breakfast in the Classroom Program on School Outcomes. JAMA Pediatrics, 169,

71–77.

Bartfeld, J., & Kim, M. 2010. Participation in the School Breakfast Program: New evidence from

the ECLS-K. Social Service Review, 84, 541–62.

Basch, C.E. 2011. Breakfast and the achievement gap among urban minority youth. Journal of

School Health, 81, 635–40.

Baxter, S.D., Hardin, J.W., Guinn, C.H., Royer, J.A., Mackelprang, A.J. & Devlin, C.M. 2010.

Children’s body mass index, participation in school meals, and observed energy intake at school

meals. International Journal of Behavioral Nutrition and Physical Activity, 7, 1–8.

Bernstein, L.S., McLaughlin, J.E., Crepinsek, M.K., Daft, L.M. & Murphy, J.M. 2004.

Evaluation of the School Breakfast Program pilot project: Final report, Nutrition Assistance

Program Report Series, No. CN-04-SBP, Project Officer: Anita Singh. U.S. Department of

Agriculture, Food and Nutrition Service, Office of Analysis, Nutrition, and Evaluation,

Alexandria, VA.

Bhattacharya, J., Currie, J. & Haider, S.J. 2006. Breakfast of champions? The School Breakfast

Program and the nutrition of children and families. Journal of Human Resources, 41, 445–66.

Briefel, R., Murphy, J.M., Kung, S., & Devaney, B. 1999. Universal-free school breakfast

program evaluation design project: Review of literature on breakfast and learning. Princeton, NJ:

Mathematica Policy Research.

Cole, T.J., Faith, M.S., Pietrobelli, A., & Heo, M. 2005. What is the best measure of adiposity

change in growing children: BMI, BMI %, BMI z-score or BMI centile? European Journal of

Clinical Nutrition, 59, 419–25.

Dahl, M.W., & Scholz, J.K. 2011. The National School Lunch Program and School Breakfast

Program: Evidence on participation and noncompliance.” Working Paper, University of

Wisconsin—Madison. Retrieved from http://www.econ.wisc.edu/~scholz/Research/Lunch.pdf.

Dotter, D. 2012. Breakfast at the desk: The impact of universal breakfast programs on academic

performance. Working Paper, University of California San Diego Department of Economics.

Retrieved from http://econweb.ucsd.edu/~ddotter/pdfs/Dotter_JMP_Manuscript.pdf

Dunifon, R. & Kowaleski-Jones, L. 2003. The influences of participation in the National School

Lunch Program and food insecurity on child well-being. Social Service Review, 77, 72–92.

34 | P a g e

Elbel, B., Corcoran, S.P., & Schwartz, A.E. 2013. Neighborhoods, schools and obesity in New

York City: The potential for place-based approaches to reduce childhood obesity. Working

Paper, New York University Institute for Education and Social Policy.

Figlio, D.N. & Winicki, J. 2005. Food for thought: The effects of school accountability plans on

school nutrition. Journal of Public Economics 89, 381–94.

FRAC. 2012. School breakfast scorecard: School year 2010-2011.

Hinrichs, P. 2010. The effects of the National School Lunch Program on education and health.

Journal of Policy Analysis and Management, 29, 479–505.

Hoyland, A., Dye, L., & Lawton, C.L. 2009. A systematic review of the effect of breakfast on

the cognitive performance of children and adolescents. Nutrition Research Reviews, 22, 220–43.

Imberman, S.A., & Kugler, A.D. 2014. The effect of providing breakfast in class on student

performance. Journal of Policy Analysis and Management, 33, 669–99.

Leos-Urbel, J. Schwartz, A.E., Weinstein, M., & Corcoran, S.P. 2013. Not just for poor kids: The

impact of universal free school breakfast on meal participation and student outcomes. Economics

of Education Review, 36, 88–107.

McEwan, P.J. 2013. The impact of Chile’s school feeding program on education outcomes.

Economics of Education Review, 32, 122–39.

Millimet, D.L., Tchernis, R., & Husain, M. 2010. School nutrition programs and the incidence of

childhood obesity. Journal of Human Resources, 45, 640–54.

Mooney, S.J., Baecker, A., & Rundle, A.G. 2013. Comparison of anthropometric and body

composition measures as predictors of components of the metabolic syndrome in a clinical

setting. Obesity Research & Clinical Practice, 7, e55–e66.

Murphy, J.M., Drake, J.E., & Weineke, K.M. 2005. Academics & breakfast connection pilot:

Final report on New York’s classroom breakfast project. Albany, NY. Retrieved from

http://hungersolutionsny.org/documents/FinalABCupdated.pdf

Pollitt, E., & Mathews, R. 1998. Breakfast and cognition: An integrative summary. The

American Journal of Clinical Nutrition, 67, 804S–813S.

Ponza, M. 1999. Universal-free school breakfast program evaluation design project: Final

evaluation design. Princeton, NJ: Mathematica Policy Research. Retrieved from

http://www.fns.usda.gov/ora/menu/DemoProjects/sbppilot/SBPdesign.PDF.

Poppendieck, J. 2010. Free for all: Fixing school food in America. Berkeley: University of

California Press.

35 | P a g e

Rampersaud, G.C., Pereira, M.A., Girard, B.L., Adams, J. & Metzl, J.D. 2005. Breakfast habits,

nutritional status, body weight, and academic performance in children and adolescents. Journal

of the American Dietetic Association, 105, 743–60.

Rundle, A., Richards, C., Bader, M.D.M., Schwartz-Soicher, O., Lee, K.K., Quinn, J., & Lovasi,

G.S. 2012. Individual- and school-level sociodemographic predictors of obesity among New

York City public school children. American Journal of Epidemiology, 176, 986–94.

Schanzenbach, D.W. 2009. Do school lunches contribute to childhood obesity? Journal of

Human Resources 44, 684–709.

Schanzenbach, D.W. & Zaki, M. 2014. Expanding the School Breakfast Program: Impacts on

children’s consumption, nutrition and health. National Bureau of Economic Research Working

Paper No. 20308.

Simeon, D.T., & Grantham-McGregor, S. 1989. Effects of missing breakfast on the cognitive

functions of school children of differing nutritional status. American Journal of Clinical

Nutrition, 49, 646–53.

Vaisman, N., Akivis, A., Vakil E., & Voet, H. 1996. Effect of breakfast timing on the cognitive

functions of elementary school students. Archives of Pediatrics & Adolescent Medicine 150,

1089–92.

Van Wye, G., Seoh, H., Adjoian, T., and Dowell, D. 2013. Evaluation of the New York City

breakfast in the classroom program. American Journal of Public Health, 103, e59–e64.

Wesnes, K.A., Pincock, C., Richardson, D., Helm, G. & Hails, S. 2003. Breakfast reduces

declines in attention and memory over the morning in schoolchildren. Appetite 41, 329–31.

Wyon, D.P., Abrahamsson, L., Järtelius, M., & Fletcher, R.G. 1997. An experimental study of

the effects of energy intake at breakfast on the test performance of 10-year-old children in

school. International Journal of Food Sciences and Nutrition, 48, 5–12.

36 | P a g e

Figure 1: Cumulative Breakfast in the Classroom Adoptions by Month, New York City

Notes: reflects all schools adopting BIC prior to June 30, 2012 that offered BIC to any of the grades K-8. Only regular public schools are included; private, charter, alternative, and special education (District 75) schools are excluded, as are suspension or other special programs.

37 | P a g e

Figure 2: Estimated Impact of BIC on Annual Breakfast and Lunch Participation Rates:

Elementary and Middle Schools

Notes: Event study coefficients from regressions of annual meal participation rates on school-level covariates, school and year fixed effects, and indicators for years before and after BIC adoption. (These indicators are equal to zero for schools that never adopted BIC, and year zero is year prior to offering BIC). Dashed and dotted lines represent a robust 95% confidence interval.

38 | P a g e

Figure 3: Estimated Impact of BIC by Year, Event Study Regressions

Notes: baseline year zero is the year prior to BIC adoption.

-.0

9-.

06

-.0

3

0

.03

.06

.09

Re

lative

to

ba

se

line

ye

ar

-2+ -1 0 1 2 3+

Math Grades 4-5

-.0

9-.

06

-.0

3

0

.03

.06

.09

Re

lative

to

ba

se

line

ye

ar

-2+ -1 0 1 2 3+

Math Grades 6-8

-.0

9-.

06

-.0

3

0

.03

.06

.09

Re

lative

to

ba

se

line

ye

ar

-2+ -1 0 1 2 3+

ELA Grades 4-5

-.0

9-.

06

-.0

3

0

.03

.06

.09

Re

lative

to

ba

se

line

ye

ar

-2+ -1 0 1 2 3+

ELA Grades 6-8

-.1

2-.

09

-.0

6-.

03

0

.03

.06

.09

Re

lative

to

ba

se

line

ye

ar

-2+ -1 0 1 2 3+

zBMI Grades K-5

-.1

5-.

1-.

05

0

.05

.1.1

5

Re

lative

to

ba

se

line

ye

ar

-2+ -1 0 1 2 3+

zBMI Grades 6-8

-.0

6-.

04

-.0

2

0

.02

.04

Re

lative

to

ba

se

line

ye

ar

-2+ -1 0 1 2 3+

Obesity Grades K-5

-.0

6-.

04

-.0

2

0

.02

.04

Re

lative

to

ba

se

line

ye

ar

-2+ -1 0 1 2 3+

Obesity Grades 6-8

-.0

1-.

00

5

0

.00

5.0

1

Re

lative

to

ba

se

line

ye

ar

-2+ -1 0 1 2 3+

>25% BIC 95% CI

Attendance Grades K-5

-.0

1-.

00

5

0

.00

5.0

1

Re

lative

to

ba

se

line

ye

ar

-2+ -1 0 1 2 3+

>25% BIC 95% CI

Attendance Grades 6-8

39 | P a g e

Table 1: Mean school characteristics by BIC adoption status and classroom coverage,

baseline

Never BIC Ever BIC

Ever BIC w/>=25% coverage

Ever full school BIC

Breakfast participation rate 24.5 23.3 23.9 24.6 Lunch participation rate 76.5 79.7 82.6 82.4 BMI z-score 0.008 0.082 0.102 0.147 Percent obese 20.8 22.5 23.7 24.9 Reading z-score (grades 3-8) 0.017 -0.128 -0.258 -0.283 Math z-score (grades 3-8) 0.002 -0.158 -0.292 -0.310 Attendance rate 92.1 91.1 90.1 91.2 Percent eligible for free lunch 67.3 74.1 81.4 80.6 Percent ELL 11.7 12.8 12.7 14.5 Percent special education 13.9 14.6 15.4 17.1 Percent Asian 12.4 8.3 3.7 4.8 Percent black 33.5 37.1 40.4 34.6 Percent Hispanic 38.5 43.9 51.5 55.9 Percent white 14.7 9.7 3.3 3.5 Percent male 50.7 50.9 50.1 50.0 Percent enrollment grades K-5 60.1 58.9 61.3 65.1 Percent in Brooklyn 33.2 28.5 23.3 7.9 Percent in Manhattan 17.3 20.6 25.9 42.1 Percent in Queens 24.3 16.7 7.8 10.5 Percent in Staten Island 4.8 4.3 0.0 0.0 Percent in Bronx 20.3 29.9 43.1 39.5 Percent UFM school 45.1 42.5 45.6 45.9 School starting time 8:20 a.m. 8:19 a.m. 8:21 a.m. 8:19 a.m. Total enrollment 657 724 683 660 N (observed in 2007) 807 281 116 38

Notes: only regular public schools serving any of the elementary and middle grades are included in the above. All means are for the 2006-07 academic year, and thus are prior to schools’ adoption of BIC. “Never BIC” refers to schools that had not adopted BIC as of June 30, 2012. “Ever BIC” refers to any school that adopted BIC prior to June 30, 2012. “Ever BIC with >=25% Coverage” refers to any school that adopted BIC prior to June 30, 2012 and offered BIC to at least 25% of classrooms. “Ever Full BIC” refers to any school that adopted BIC school-wide prior to June 30, 2012. In the few cases where BIC coverage changed over time, we classified schools according to their highest extent of coverage.

40 | P a g e

Table 2: Impact of BIC adoption on meals program participation, 2001 – 2012

All schools Elementary Middle

Breakfast: (1) Post BIC adoption 0.118*** 0.130*** 0.130***

(0.009) (0.012) (0.020) (2) Post BIC adoption: school 0.211*** 0.231*** 0.251*** with >25% coverage (0.016) (0.020) (0.031) (3) Post BIC adoption: full school 0.290*** 0.319*** 0.319***

(0.032) (0.044) (0.032)

Lunch: (4) Post BIC adoption -0.001 0.008 -0.014 (0.005) (0.005) (0.014) (5) Post BIC adoption: school -0.008 0.007 -0.014 with >25% coverage (0.008) (0.007) (0.020) (6) Post BIC adoption: full school -0.005 -0.002 0.048

(0.018) (0.018) (0.048)

N – breakfast (school x year) 12,407 7,833 2,598 Mean breakfast participation pre-2008 0.201 0.236 0.120

N – lunch (school x year) 12,062 7,518 2,565 Mean lunch participation 2008 0.732 0.813 0.629

Notes: each cell is a coefficient estimate from a separate regression model. In rows (1) and (4), the coefficient is for

the “post BIC adoption” indicator equal to one in school-years where BIC was offered in the school. For rows (2)

and (4), this indicator is equal to one only if BIC was offered to at least 25% of classrooms in the school. For rows

(3) and (6), this indicator is equal to one only if BIC was offered school-wide. The columns represent subsamples:

all schools, elementary schools only (including elementary/ middle combinations), and middle schools only

(including middle/high combinations). The dependent variable is the annual breakfast or lunch participation rate

for a given school and year, measured as average daily meals served divided by average daily attendance. Standard

errors, robust to clustering within schools, are in parentheses (*** p<0.001, ** p<0.01, * p<0.05).

41 | P a g e

Table 3: Impact of BIC on obesity and BMI

Pre-post Cumulative days Grade K-5 Grade 6-8 Grade K-5 Grade 6-8

Panel A: Impact on zBMI (1) Post BIC adoption -0.0036 -0.0239 0.0030 0.0064 (0.0114) (0.0184) (0.0044) (0.0041) (2) Post BIC adoption: school -0.0076 -0.0147 -0.0044 0.0079 with >25% coverage (0.0157) (0.0320) (0.0070) (0.0067) (3) Post BIC adoption: full school -0.0156 -0.0225 -0.0243 0.0024 (0.0315) (0.0161) (0.0176) (0.0088) Panel B: Impact on obesity (4) Post BIC adoption -0.0007 -0.0087 0.0020 0.0026 (0.0030) (0.0070) (0.0013) (0.0015) (5) Post BIC adoption: school 0.0004 -0.0074 0.0013 0.0027 with >25% coverage (0.0045) (0.0125) (0.0021) (0.0026) (6) Post BIC adoption: full school -0.0024 -0.0020 -0.0035 0.0010 (0.0089) (0.0059) (0.0059) (0.0033) Observations 2,131,469 980,088 2,131,469 980,088

Notes: Standard errors in parentheses, robust to clustering at the school level. Obese is defined as being above the 95th percentile nationally for one’s gender

and age in months, based on the 2000 CDC BMI-for-age charts. All models include student covariates, grade, school, and year effects. Covariates include age,

gender, race/ethnicity, low income status, LEP, immigrant, and special education status. Low income is measured by eligibility for free or reduced price meals

or enrollment in a Universal Free Meal school. Age is measured in months at the time of the Fitnessgram. We exclude charter school students, students

attending citywide special education schools (District 75), students in schools where Fitnessgram coverage is less than 50 percent, and students with

biologically implausible BMIs.

42 | P a g e

Table 4: Impact of BIC on ELA and math achievement

Grades 4-5 Pre-post

Grades 6-8 Pre-post

Grades 4-5 Cumulative days

Grades 6-8 Cumulative days

ELA Math ELA Math ELA Math ELA Math

(1) Post BIC adoption -0.011 -0.005 -0.000 0.007 -0.003 -0.001 -0.002 0.005 (0.008) (0.011) (0.008) (0.011) (0.002) (0.004) (0.002) (0.003) (2) Post BIC adoption: school -0.021 -0.005 -0.020* 0.005 -0.006 -0.001 -0.005* 0.008* with >25% coverage (0.014) (0.018) (0.009) (0.017) (0.003) (0.004) (0.002) (0.004) (3) Post BIC adoption: full school -0.041 0.028 -0.015 -0.008 -0.007 0.014 -0.005 0.012 (0.027) (0.037) (0.019) (0.034) (0.009) (0.010) (0.004) (0.009) Observations 717,486 744,934 1,097,593 1,126,258 717,486 744,934 1,097,593 1,126,258

Notes: Standard errors robust to clustering at the school level shown in parentheses (* p<0.05). All Models control for lagged z-score, race/ethnicity, low

income status, LEP, immigrant, and special education status. Low income is measured by eligibility for free or reduced price meals or enrollment in a Universal

Free Meal school. Excludes charter school students and students attending citywide special education schools (District 75).

43 | P a g e

Table 5: Impact of BIC on attendance

Pre-post Cumulative days Grade K-5 Grade 6-8 Grade K-5 Grade 6-8

(1) Post BIC adoption <0.001 0.003* <0.001 0.001* (0.001) (0.001) (<0.001) (0.000) (2) Post BIC adoption: school 0.001 0.004* <0.001 0.001 with >25% coverage (0.001) (0.002) (<0.001) (0.001)

(3) Post BIC adoption: full school <0.001 0.004 <0.001 <0.001 (0.001) (0.003) (0.001) (0.001) Observations 2,496,321 1,228,769 2,496,321 1,228,769

Notes: Standard errors robust to clustering at the school level shown in parentheses (* p<0.05). All Models control for race/ethnicity, low income status, LEP,

immigrant, and special education status. Low income is measured by eligibility for free or reduced price meals or enrollment in a Universal Free Meal school.

Excludes charter school students and students attending citywide special education schools (District 75).